CV72AQ - Ambarella launches next-generation automotive-grade 5nm process AI SoC

Single-chip navigation and 8-megapixel front-view ADAS to run Transformer efficiently

April 17, 2023, Shanghai, China— Ambarella (hereinafter referred to as "Ambarella", a semiconductor company focusing on AI visual perception chips) today announced the launch of an AI vision system-level chip based on Ambarella's new generation CVflow®3.0 AI architecture ( SoC) CV72AQ, targeting the automotive application market. Through the latest CVflowTM architecture, the AI ​​performance of CV72AQ is 6 times higher than that of the previous generation product CV22AQ under the same power consumption, and it can also efficiently run the latest deep learning algorithm based on Transformer neural network. Efficient AI performance allows CV72AQ to simultaneously support the fusion of cameras, millimeter wave radar and ultrasonic radar. Combined with rich functions such as high-performance ISP image processing and H265 encoding, CV72AQ is an ideal platform for solutions such as forward-looking ADAS all-in-one machine and single-chip 6V5R "all-in-one parking and parking" solution.

In the CV72AQ, Ambarella follows the "algorithm first" design concept and tailors the CVflow 3.0 architecture to achieve the best AI performance per unit of energy consumption. CV72AQ integrates Ambarella's new generation ISP, which can provide excellent high-definition images under extremely low light conditions, continuing to maintain Ambarella's leading position in the image signal processing industry. CV72AQ has a high-efficiency video encoder that supports image processing and simultaneous video encoding up to 8MP90 (720MP/s), which is 3 times faster than the previous generation CV22AQ. The CPU performance of CV72AQ SoC is also 2 times higher than that of the previous generation CV22AQ. It can provide a single-chip solution for forward-looking ADAS and multi-camera integrated parking systems, including camera and radar integration, which not only reduces the overall BOM cost, but also reduces the software complexity. Complexity.

Ambarella President and CEO Wang Fengmin said: "Our latest SoC CV72AQ provides the industry with higher AI performance and throughput without increasing power consumption, and continues to accelerate innovation in the automotive market. Through 5 nanometer automotive regulations Process, and the third generation of CVFlow, achieve higher chip integration, processing efficiency, and higher AI performance, making it possible to design a domain controller on a single chip.”

Ambarella uses proprietary Oculii™ Virtual Aperture Imaging (VAI) AI radar technology to integrate hardware acceleration into the CV72AQ to achieve fusion with camera image data and enable dynamic control between the radar and camera. Ambarella's radar technology can improve resolution by 10 to 100 times compared to other traditional radar solutions currently on the market. These radars and integrated AI processing capabilities can support automotive perception systems that can operate normally even in severe weather conditions such as rain, snow, and fog, and can enhance image data from cameras in dark environments at night.

Ambarella CVflow AI development platform provides a set of easy-to-use algorithm development and optimization tools, supporting all common machine learning frameworks. With its rich integration functions and excellent compatibility, this set of tools supports all Ambarella CV series AI SoCs. Through this set of AI tools, CV72AQ further reduces product design and development costs and maximizes software reusability.

CV72AQ in action

Supports forward-looking ADAS with resolutions up to 8MP30, supports 1V1R visual perception and millimeter-wave radar fusion, and can be attached with a driver monitoring system DMS. The system has ultra-low power consumption and is suitable for all-in-one solutions installed on the front windshield.

Supports 5V5R/6V6R ADAS + parking “in-one parking” and 5V5R/6V5R multi-sensor data fusion.

Supports efficient ISP processing and H265 encoding of input data from multiple cameras at the same time, up to 8MP90 (720MP/s).

A single chip supports the deployment of various AI algorithms required for L2+ integrated travel products.